On 14-Sep-06, at 9:45 AM, Daniel Stenning wrote:
Naturally one could just use the MemoryBlock built in methods to
address the
block directly, but this currently requires the developer to calculate
offsets himself.
Another option would be for MemoryBlocks to access data by [index]
( like C
) instead of by byte offset. For integer, floats, and structures
this
should be doable.
You can do this yourself with a memoryblock subclass of course, and
if you use a cached Ptr to avoid the memoryblock accessors this
approach isn't much slower than using the built-in accessors
directly. Just keep in mind the Ptr has to be refreshed when ever the
block is reallocated (you can overload the Size method to do this
automatically).
When you need speed instead of pretty code just access the pointer
directly and calculate the offsets - I'll often define a stride
constant in the memoryblock subclass to keep things manageable. With
this in mind, pretty code for an array of single-precision floats
might look like this:
m.Float(3) = 1.0 // 1 virtual function call
When you need speed it looks more like:
m.Pointer( m.kStride * 3 ) = 1.0 // No function calls!
Not pretty, but it's much faster (though sadly still not as fast as C
code) and if you decide you need doubles later on just change the
stride and everything will continue to work. You can also pass
m.Pointer to declares to save a conversion during those calls.
I agree it would be nice to be able to define a byte stride (and even
interleaved strides) for MemoryBlock and Ptr to avoid these extra
steps but it would have to be implemented at the preprocessor or
compiler level to be any good. I'd be happy with a more robust Ptr
type - e.g. instead of "Dim p As Ptr" use something like "Pointer p
As Integer" or "Pointer p As myStruct" where p's offsets match it's
type.
Frank.
<http://developer.chaoticbox.com/>
_______________________________________________
Unsubscribe or switch delivery mode:
<http://www.realsoftware.com/support/listmanager/>
Search the archives of this list here:
<http://support.realsoftware.com/listarchives/lists.html>
